This lecture introduces logistic regression as a model for binary classification that predicts probabilities. We will see how the naive Bayes model and the logistic regression model are intimately connected with each other.
The lecture then proceeds to the question of training a logistic regression classifier using the maximum likelihood and the maximum a posteriori criteria. We will end with a discussion that connects the logistic regression classifier with the general idea of learning by loss minimization.
Links and Resources
Chapter 6 of Tom Mitchell’s book
Chapter 7 of A Course in Machine Learning by Hal Daume
Chapters 2 and 3 of Pattern Classification by Duda, Hart and Stork